BRAHMa: a YOLO detector measures the bars in spiral galaxies almost as accurately as two catalogues combined

Trained on annotations from Galaxy Zoo 3D and images from the DESI Legacy Survey, this publicly available tool is designed for the major surveys to come, which manual catalogues can no longer keep up with.

The central bars of spiral galaxies are a key object of study for dark matter and galactic evolution, but measuring them remains laborious: catalogues compiled by hand or by volunteers will not keep pace with the next major surveys, and existing machine learning approaches, based on pixel-by-pixel segmentation, are computationally intensive and require complex post-processing to extract physical parameters.

Rajit Shrivastava, Narendra Nath Patra and Keerthi K present BRAHMa, a tool that directly frames each bar in an oriented box using a YOLO11x model. Validated first on synthetic data, it was then trained on the contours of bars annotated by the Galaxy Zoo 3D citizen science project, using images from the DESI Legacy Survey.

Closer to the data limit

In the absence of absolute ground truth, the authors compare two catalogues on 586 common galaxies and obtain an average deviation of 0.99 kiloparsec, which they take as a reference. Out of 3,150 galaxies in the Hoyle dataset, BRAHMa achieves an error of 1.1 kiloparsec, close to this limit, and is robust on other datasets. The tool is publicly accessible via a web interface.